net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 unsigned int ovs_net_id __read_mostly;
  62
  63 static struct genl_family dp_packet_genl_family;
  64 static struct genl_family dp_flow_genl_family;
  65 static struct genl_family dp_datapath_genl_family;
  66
  67 static const struct nla_policy flow_policy[];
  68
  69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP,
  71 };
  72
  73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP,
  75 };
  76
  77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP,
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  84                             unsigned int group)
  85 {
  86         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  87                genl_has_listeners(family, genl_info_net(info), group);
  88 }
  89
  90 static void ovs_notify(struct genl_family *family,
  91                        struct sk_buff *skb, struct genl_info *info)
  92 {
  93         genl_notify(family, skb, info, 0, GFP_KERNEL);
  94 }
  95
  96 /**
  97  * DOC: Locking:
  98  *
  99  * All writes e.g. Writes to device state (add/remove datapath, port, set
 100  * operations on vports, etc.), Writes to other state (flow table
 101  * modifications, set miscellaneous datapath parameters, etc.) are protected
 102  * by ovs_lock.
 103  *
 104  * Reads are protected by RCU.
 105  *
 106  * There are a few special cases (mostly stats) that have their own
 107  * synchronization but they nest under all of above and don't interact with
 108  * each other.
 109  *
 110  * The RTNL lock nests inside ovs_mutex.
 111  */
 112
 113 static DEFINE_MUTEX(ovs_mutex);
 114
 115 void ovs_lock(void)
 116 {
 117         mutex_lock(&ovs_mutex);
 118 }
 119
 120 void ovs_unlock(void)
 121 {
 122         mutex_unlock(&ovs_mutex);
 123 }
 124
 125 #ifdef CONFIG_LOCKDEP
 126 int lockdep_ovsl_is_held(void)
 127 {
 128         if (debug_locks)
 129                 return lockdep_is_held(&ovs_mutex);
 130         else
 131                 return 1;
 132 }
 133 #endif
 134
 135 static struct vport *new_vport(const struct vport_parms *);
 136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 137                              const struct sw_flow_key *,
 138                              const struct dp_upcall_info *,
 139                              uint32_t cutlen);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *,
 142                                   const struct dp_upcall_info *,
 143                                   uint32_t cutlen);
 144
 145 /* Must be called with rcu_read_lock. */
 146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 147 {
 148         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 149
 150         if (dev) {
 151                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 152                 if (vport)
 153                         return vport->dp;
 154         }
 155
 156         return NULL;
 157 }
 158
 159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 160  * returned dp pointer valid.
 161  */
 162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 163 {
 164         struct datapath *dp;
 165
 166         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 167         rcu_read_lock();
 168         dp = get_dp_rcu(net, dp_ifindex);
 169         rcu_read_unlock();
 170
 171         return dp;
 172 }
 173
 174 /* Must be called with rcu_read_lock or ovs_mutex. */
 175 const char *ovs_dp_name(const struct datapath *dp)
 176 {
 177         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 178         return ovs_vport_name(vport);
 179 }
 180
 181 static int get_dpifindex(const struct datapath *dp)
 182 {
 183         struct vport *local;
 184         int ifindex;
 185
 186         rcu_read_lock();
 187
 188         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 189         if (local)
 190                 ifindex = local->dev->ifindex;
 191         else
 192                 ifindex = 0;
 193
 194         rcu_read_unlock();
 195
 196         return ifindex;
 197 }
 198
 199 static void destroy_dp_rcu(struct rcu_head *rcu)
 200 {
 201         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 202
 203         ovs_flow_tbl_destroy(&dp->table);
 204         free_percpu(dp->stats_percpu);
 205         kfree(dp->ports);
 206         kfree(dp);
 207 }
 208
 209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 210                                             u16 port_no)
 211 {
 212         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 213 }
 214
 215 /* Called with ovs_mutex or RCU read lock. */
 216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 217 {
 218         struct vport *vport;
 219         struct hlist_head *head;
 220
 221         head = vport_hash_bucket(dp, port_no);
 222         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 223                 if (vport->port_no == port_no)
 224                         return vport;
 225         }
 226         return NULL;
 227 }
 228
 229 /* Called with ovs_mutex. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = ovs_vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 238
 239                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 240         }
 241         return vport;
 242 }
 243
 244 void ovs_dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_OVSL();
 247
 248         /* First drop references to device. */
 249         hlist_del_rcu(&p->dp_hash_node);
 250
 251         /* Then destroy it. */
 252         ovs_vport_del(p);
 253 }
 254
 255 /* Must be called with rcu_read_lock. */
 256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 257 {
 258         const struct vport *p = OVS_CB(skb)->input_vport;
 259         struct datapath *dp = p->dp;
 260         struct sw_flow *flow;
 261         struct sw_flow_actions *sf_acts;
 262         struct dp_stats_percpu *stats;
 263         u64 *stats_counter;
 264         u32 n_mask_hit;
 265
 266         stats = this_cpu_ptr(dp->stats_percpu);
 267
 268         /* Look up flow. */
 269         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272                 int error;
 273
 274                 memset(&upcall, 0, sizeof(upcall));
 275                 upcall.cmd = OVS_PACKET_CMD_MISS;
 276                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 277                 upcall.mru = OVS_CB(skb)->mru;
 278                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 279                 if (unlikely(error))
 280                         kfree_skb(skb);
 281                 else
 282                         consume_skb(skb);
 283                 stats_counter = &stats->n_missed;
 284                 goto out;
 285         }
 286
 287         ovs_flow_stats_update(flow, key->tp.flags, skb);
 288         sf_acts = rcu_dereference(flow->sf_acts);
 289         ovs_execute_actions(dp, skb, sf_acts, key);
 290
 291         stats_counter = &stats->n_hit;
 292
 293 out:
 294         /* Update datapath statistics. */
 295         u64_stats_update_begin(&stats->syncp);
 296         (*stats_counter)++;
 297         stats->n_mask_hit += n_mask_hit;
 298         u64_stats_update_end(&stats->syncp);
 299 }
 300
 301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 302                   const struct sw_flow_key *key,
 303                   const struct dp_upcall_info *upcall_info,
 304                   uint32_t cutlen)
 305 {
 306         struct dp_stats_percpu *stats;
 307         int err;
 308
 309         if (upcall_info->portid == 0) {
 310                 err = -ENOTCONN;
 311                 goto err;
 312         }
 313
 314         if (!skb_is_gso(skb))
 315                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 316         else
 317                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 318         if (err)
 319                 goto err;
 320
 321         return 0;
 322
 323 err:
 324         stats = this_cpu_ptr(dp->stats_percpu);
 325
 326         u64_stats_update_begin(&stats->syncp);
 327         stats->n_lost++;
 328         u64_stats_update_end(&stats->syncp);
 329
 330         return err;
 331 }
 332
 333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 334                              const struct sw_flow_key *key,
 335                              const struct dp_upcall_info *upcall_info,
 336                                  uint32_t cutlen)
 337 {
 338         struct sk_buff *segs, *nskb;
 339         int err;
 340
 341         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 342         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 343         if (IS_ERR(segs))
 344                 return PTR_ERR(segs);
 345         if (segs == NULL)
 346                 return -EINVAL;
 347
 348         /* Queue all of the segments. */
 349         skb = segs;
 350         do {
 351                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 352                 if (err)
 353                         break;
 354
 355         } while ((skb = skb->next));
 356
 357         /* Free all of the segments. */
 358         skb = segs;
 359         do {
 360                 nskb = skb->next;
 361                 if (err)
 362                         kfree_skb(skb);
 363                 else
 364                         consume_skb(skb);
 365         } while ((skb = nskb));
 366         return err;
 367 }
 368
 369 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 370                               unsigned int hdrlen)
 371 {
 372         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 373                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 374                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 375                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 376
 377         /* OVS_PACKET_ATTR_USERDATA */
 378         if (upcall_info->userdata)
 379                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 380
 381         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 382         if (upcall_info->egress_tun_info)
 383                 size += nla_total_size(ovs_tun_key_attr_size());
 384
 385         /* OVS_PACKET_ATTR_ACTIONS */
 386         if (upcall_info->actions_len)
 387                 size += nla_total_size(upcall_info->actions_len);
 388
 389         /* OVS_PACKET_ATTR_MRU */
 390         if (upcall_info->mru)
 391                 size += nla_total_size(sizeof(upcall_info->mru));
 392
 393         return size;
 394 }
 395
 396 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 397 {
 398         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 399                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 400
 401                 if (plen > 0)
 402                         skb_put_zero(skb, plen);
 403         }
 404 }
 405
 406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 407                                   const struct sw_flow_key *key,
 408                                   const struct dp_upcall_info *upcall_info,
 409                                   uint32_t cutlen)
 410 {
 411         struct ovs_header *upcall;
 412         struct sk_buff *nskb = NULL;
 413         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 414         struct nlattr *nla;
 415         size_t len;
 416         unsigned int hlen;
 417         int err, dp_ifindex;
 418
 419         dp_ifindex = get_dpifindex(dp);
 420         if (!dp_ifindex)
 421                 return -ENODEV;
 422
 423         if (skb_vlan_tag_present(skb)) {
 424                 nskb = skb_clone(skb, GFP_ATOMIC);
 425                 if (!nskb)
 426                         return -ENOMEM;
 427
 428                 nskb = __vlan_hwaccel_push_inside(nskb);
 429                 if (!nskb)
 430                         return -ENOMEM;
 431
 432                 skb = nskb;
 433         }
 434
 435         if (nla_attr_size(skb->len) > USHRT_MAX) {
 436                 err = -EFBIG;
 437                 goto out;
 438         }
 439
 440         /* Complete checksum if needed */
 441         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 442             (err = skb_csum_hwoffload_help(skb, 0)))
 443                 goto out;
 444
 445         /* Older versions of OVS user space enforce alignment of the last
 446          * Netlink attribute to NLA_ALIGNTO which would require extensive
 447          * padding logic. Only perform zerocopy if padding is not required.
 448          */
 449         if (dp->user_features & OVS_DP_F_UNALIGNED)
 450                 hlen = skb_zerocopy_headlen(skb);
 451         else
 452                 hlen = skb->len;
 453
 454         len = upcall_msg_size(upcall_info, hlen - cutlen);
 455         user_skb = genlmsg_new(len, GFP_ATOMIC);
 456         if (!user_skb) {
 457                 err = -ENOMEM;
 458                 goto out;
 459         }
 460
 461         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 462                              0, upcall_info->cmd);
 463         upcall->dp_ifindex = dp_ifindex;
 464
 465         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 466         BUG_ON(err);
 467
 468         if (upcall_info->userdata)
 469                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 470                           nla_len(upcall_info->userdata),
 471                           nla_data(upcall_info->userdata));
 472
 473         if (upcall_info->egress_tun_info) {
 474                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 475                 err = ovs_nla_put_tunnel_info(user_skb,
 476                                               upcall_info->egress_tun_info);
 477                 BUG_ON(err);
 478                 nla_nest_end(user_skb, nla);
 479         }
 480
 481         if (upcall_info->actions_len) {
 482                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 483                 err = ovs_nla_put_actions(upcall_info->actions,
 484                                           upcall_info->actions_len,
 485                                           user_skb);
 486                 if (!err)
 487                         nla_nest_end(user_skb, nla);
 488                 else
 489                         nla_nest_cancel(user_skb, nla);
 490         }
 491
 492         /* Add OVS_PACKET_ATTR_MRU */
 493         if (upcall_info->mru) {
 494                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 495                                 upcall_info->mru)) {
 496                         err = -ENOBUFS;
 497                         goto out;
 498                 }
 499                 pad_packet(dp, user_skb);
 500         }
 501
 502         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 503         if (cutlen > 0) {
 504                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 505                                 skb->len)) {
 506                         err = -ENOBUFS;
 507                         goto out;
 508                 }
 509                 pad_packet(dp, user_skb);
 510         }
 511
 512         /* Only reserve room for attribute header, packet data is added
 513          * in skb_zerocopy() */
 514         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 515                 err = -ENOBUFS;
 516                 goto out;
 517         }
 518         nla->nla_len = nla_attr_size(skb->len - cutlen);
 519
 520         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 521         if (err)
 522                 goto out;
 523
 524         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 525         pad_packet(dp, user_skb);
 526
 527         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 528
 529         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 530         user_skb = NULL;
 531 out:
 532         if (err)
 533                 skb_tx_error(skb);
 534         kfree_skb(user_skb);
 535         kfree_skb(nskb);
 536         return err;
 537 }
 538
 539 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 540 {
 541         struct ovs_header *ovs_header = info->userhdr;
 542         struct net *net = sock_net(skb->sk);
 543         struct nlattr **a = info->attrs;
 544         struct sw_flow_actions *acts;
 545         struct sk_buff *packet;
 546         struct sw_flow *flow;
 547         struct sw_flow_actions *sf_acts;
 548         struct datapath *dp;
 549         struct vport *input_vport;
 550         u16 mru = 0;
 551         int len;
 552         int err;
 553         bool log = !a[OVS_PACKET_ATTR_PROBE];
 554
 555         err = -EINVAL;
 556         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 557             !a[OVS_PACKET_ATTR_ACTIONS])
 558                 goto err;
 559
 560         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 561         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 562         err = -ENOMEM;
 563         if (!packet)
 564                 goto err;
 565         skb_reserve(packet, NET_IP_ALIGN);
 566
 567         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 568
 569         /* Set packet's mru */
 570         if (a[OVS_PACKET_ATTR_MRU]) {
 571                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 572                 packet->ignore_df = 1;
 573         }
 574         OVS_CB(packet)->mru = mru;
 575
 576         /* Build an sw_flow for sending this packet. */
 577         flow = ovs_flow_alloc();
 578         err = PTR_ERR(flow);
 579         if (IS_ERR(flow))
 580                 goto err_kfree_skb;
 581
 582         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 583                                              packet, &flow->key, log);
 584         if (err)
 585                 goto err_flow_free;
 586
 587         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 588                                    &flow->key, &acts, log);
 589         if (err)
 590                 goto err_flow_free;
 591
 592         rcu_assign_pointer(flow->sf_acts, acts);
 593         packet->priority = flow->key.phy.priority;
 594         packet->mark = flow->key.phy.skb_mark;
 595
 596         rcu_read_lock();
 597         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 598         err = -ENODEV;
 599         if (!dp)
 600                 goto err_unlock;
 601
 602         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 603         if (!input_vport)
 604                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 605
 606         if (!input_vport)
 607                 goto err_unlock;
 608
 609         packet->dev = input_vport->dev;
 610         OVS_CB(packet)->input_vport = input_vport;
 611         sf_acts = rcu_dereference(flow->sf_acts);
 612
 613         local_bh_disable();
 614         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 615         local_bh_enable();
 616         rcu_read_unlock();
 617
 618         ovs_flow_free(flow, false);
 619         return err;
 620
 621 err_unlock:
 622         rcu_read_unlock();
 623 err_flow_free:
 624         ovs_flow_free(flow, false);
 625 err_kfree_skb:
 626         kfree_skb(packet);
 627 err:
 628         return err;
 629 }
 630
 631 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 632         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 633         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 634         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 635         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 636         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 637 };
 638
 639 static const struct genl_ops dp_packet_genl_ops[] = {
 640         { .cmd = OVS_PACKET_CMD_EXECUTE,
 641           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 642           .policy = packet_policy,
 643           .doit = ovs_packet_cmd_execute
 644         }
 645 };
 646
 647 static struct genl_family dp_packet_genl_family __ro_after_init = {
 648         .hdrsize = sizeof(struct ovs_header),
 649         .name = OVS_PACKET_FAMILY,
 650         .version = OVS_PACKET_VERSION,
 651         .maxattr = OVS_PACKET_ATTR_MAX,
 652         .netnsok = true,
 653         .parallel_ops = true,
 654         .ops = dp_packet_genl_ops,
 655         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 656         .module = THIS_MODULE,
 657 };
 658
 659 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 660                          struct ovs_dp_megaflow_stats *mega_stats)
 661 {
 662         int i;
 663
 664         memset(mega_stats, 0, sizeof(*mega_stats));
 665
 666         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 667         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 668
 669         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 670
 671         for_each_possible_cpu(i) {
 672                 const struct dp_stats_percpu *percpu_stats;
 673                 struct dp_stats_percpu local_stats;
 674                 unsigned int start;
 675
 676                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 677
 678                 do {
 679                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 680                         local_stats = *percpu_stats;
 681                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 682
 683                 stats->n_hit += local_stats.n_hit;
 684                 stats->n_missed += local_stats.n_missed;
 685                 stats->n_lost += local_stats.n_lost;
 686                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 687         }
 688 }
 689
 690 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 691 {
 692         return ovs_identifier_is_ufid(sfid) &&
 693                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 694 }
 695
 696 static bool should_fill_mask(uint32_t ufid_flags)
 697 {
 698         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 699 }
 700
 701 static bool should_fill_actions(uint32_t ufid_flags)
 702 {
 703         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 704 }
 705
 706 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 707                                     const struct sw_flow_id *sfid,
 708                                     uint32_t ufid_flags)
 709 {
 710         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 711
 712         /* OVS_FLOW_ATTR_UFID */
 713         if (sfid && ovs_identifier_is_ufid(sfid))
 714                 len += nla_total_size(sfid->ufid_len);
 715
 716         /* OVS_FLOW_ATTR_KEY */
 717         if (!sfid || should_fill_key(sfid, ufid_flags))
 718                 len += nla_total_size(ovs_key_attr_size());
 719
 720         /* OVS_FLOW_ATTR_MASK */
 721         if (should_fill_mask(ufid_flags))
 722                 len += nla_total_size(ovs_key_attr_size());
 723
 724         /* OVS_FLOW_ATTR_ACTIONS */
 725         if (should_fill_actions(ufid_flags))
 726                 len += nla_total_size(acts->orig_len);
 727
 728         return len
 729                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 730                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 731                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 732 }
 733
 734 /* Called with ovs_mutex or RCU read lock. */
 735 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 736                                    struct sk_buff *skb)
 737 {
 738         struct ovs_flow_stats stats;
 739         __be16 tcp_flags;
 740         unsigned long used;
 741
 742         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 743
 744         if (used &&
 745             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 746                               OVS_FLOW_ATTR_PAD))
 747                 return -EMSGSIZE;
 748
 749         if (stats.n_packets &&
 750             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 751                           sizeof(struct ovs_flow_stats), &stats,
 752                           OVS_FLOW_ATTR_PAD))
 753                 return -EMSGSIZE;
 754
 755         if ((u8)ntohs(tcp_flags) &&
 756              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 757                 return -EMSGSIZE;
 758
 759         return 0;
 760 }
 761
 762 /* Called with ovs_mutex or RCU read lock. */
 763 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 764                                      struct sk_buff *skb, int skb_orig_len)
 765 {
 766         struct nlattr *start;
 767         int err;
 768
 769         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 770          * this is the first flow to be dumped into 'skb'.  This is unusual for
 771          * Netlink but individual action lists can be longer than
 772          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 773          * The userspace caller can always fetch the actions separately if it
 774          * really wants them.  (Most userspace callers in fact don't care.)
 775          *
 776          * This can only fail for dump operations because the skb is always
 777          * properly sized for single flows.
 778          */
 779         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 780         if (start) {
 781                 const struct sw_flow_actions *sf_acts;
 782
 783                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 784                 err = ovs_nla_put_actions(sf_acts->actions,
 785                                           sf_acts->actions_len, skb);
 786
 787                 if (!err)
 788                         nla_nest_end(skb, start);
 789                 else {
 790                         if (skb_orig_len)
 791                                 return err;
 792
 793                         nla_nest_cancel(skb, start);
 794                 }
 795         } else if (skb_orig_len) {
 796                 return -EMSGSIZE;
 797         }
 798
 799         return 0;
 800 }
 801
 802 /* Called with ovs_mutex or RCU read lock. */
 803 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 804                                   struct sk_buff *skb, u32 portid,
 805                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 806 {
 807         const int skb_orig_len = skb->len;
 808         struct ovs_header *ovs_header;
 809         int err;
 810
 811         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 812                                  flags, cmd);
 813         if (!ovs_header)
 814                 return -EMSGSIZE;
 815
 816         ovs_header->dp_ifindex = dp_ifindex;
 817
 818         err = ovs_nla_put_identifier(flow, skb);
 819         if (err)
 820                 goto error;
 821
 822         if (should_fill_key(&flow->id, ufid_flags)) {
 823                 err = ovs_nla_put_masked_key(flow, skb);
 824                 if (err)
 825                         goto error;
 826         }
 827
 828         if (should_fill_mask(ufid_flags)) {
 829                 err = ovs_nla_put_mask(flow, skb);
 830                 if (err)
 831                         goto error;
 832         }
 833
 834         err = ovs_flow_cmd_fill_stats(flow, skb);
 835         if (err)
 836                 goto error;
 837
 838         if (should_fill_actions(ufid_flags)) {
 839                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 840                 if (err)
 841                         goto error;
 842         }
 843
 844         genlmsg_end(skb, ovs_header);
 845         return 0;
 846
 847 error:
 848         genlmsg_cancel(skb, ovs_header);
 849         return err;
 850 }
 851
 852 /* May not be called with RCU read lock. */
 853 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 854                                                const struct sw_flow_id *sfid,
 855                                                struct genl_info *info,
 856                                                bool always,
 857                                                uint32_t ufid_flags)
 858 {
 859         struct sk_buff *skb;
 860         size_t len;
 861
 862         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 863                 return NULL;
 864
 865         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 866         skb = genlmsg_new(len, GFP_KERNEL);
 867         if (!skb)
 868                 return ERR_PTR(-ENOMEM);
 869
 870         return skb;
 871 }
 872
 873 /* Called with ovs_mutex. */
 874 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 875                                                int dp_ifindex,
 876                                                struct genl_info *info, u8 cmd,
 877                                                bool always, u32 ufid_flags)
 878 {
 879         struct sk_buff *skb;
 880         int retval;
 881
 882         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 883                                       &flow->id, info, always, ufid_flags);
 884         if (IS_ERR_OR_NULL(skb))
 885                 return skb;
 886
 887         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 888                                         info->snd_portid, info->snd_seq, 0,
 889                                         cmd, ufid_flags);
 890         BUG_ON(retval < 0);
 891         return skb;
 892 }
 893
 894 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 895 {
 896         struct net *net = sock_net(skb->sk);
 897         struct nlattr **a = info->attrs;
 898         struct ovs_header *ovs_header = info->userhdr;
 899         struct sw_flow *flow = NULL, *new_flow;
 900         struct sw_flow_mask mask;
 901         struct sk_buff *reply;
 902         struct datapath *dp;
 903         struct sw_flow_actions *acts;
 904         struct sw_flow_match match;
 905         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 906         int error;
 907         bool log = !a[OVS_FLOW_ATTR_PROBE];
 908
 909         /* Must have key and actions. */
 910         error = -EINVAL;
 911         if (!a[OVS_FLOW_ATTR_KEY]) {
 912                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 913                 goto error;
 914         }
 915         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 916                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 917                 goto error;
 918         }
 919
 920         /* Most of the time we need to allocate a new flow, do it before
 921          * locking.
 922          */
 923         new_flow = ovs_flow_alloc();
 924         if (IS_ERR(new_flow)) {
 925                 error = PTR_ERR(new_flow);
 926                 goto error;
 927         }
 928
 929         /* Extract key. */
 930         ovs_match_init(&match, &new_flow->key, false, &mask);
 931         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 932                                   a[OVS_FLOW_ATTR_MASK], log);
 933         if (error)
 934                 goto err_kfree_flow;
 935
 936         /* Extract flow identifier. */
 937         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 938                                        &new_flow->key, log);
 939         if (error)
 940                 goto err_kfree_flow;
 941
 942         /* unmasked key is needed to match when ufid is not used. */
 943         if (ovs_identifier_is_key(&new_flow->id))
 944                 match.key = new_flow->id.unmasked_key;
 945
 946         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 947
 948         /* Validate actions. */
 949         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 950                                      &new_flow->key, &acts, log);
 951         if (error) {
 952                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 953                 goto err_kfree_flow;
 954         }
 955
 956         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 957                                         ufid_flags);
 958         if (IS_ERR(reply)) {
 959                 error = PTR_ERR(reply);
 960                 goto err_kfree_acts;
 961         }
 962
 963         ovs_lock();
 964         dp = get_dp(net, ovs_header->dp_ifindex);
 965         if (unlikely(!dp)) {
 966                 error = -ENODEV;
 967                 goto err_unlock_ovs;
 968         }
 969
 970         /* Check if this is a duplicate flow */
 971         if (ovs_identifier_is_ufid(&new_flow->id))
 972                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 973         if (!flow)
 974                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 975         if (likely(!flow)) {
 976                 rcu_assign_pointer(new_flow->sf_acts, acts);
 977
 978                 /* Put flow in bucket. */
 979                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 980                 if (unlikely(error)) {
 981                         acts = NULL;
 982                         goto err_unlock_ovs;
 983                 }
 984
 985                 if (unlikely(reply)) {
 986                         error = ovs_flow_cmd_fill_info(new_flow,
 987                                                        ovs_header->dp_ifindex,
 988                                                        reply, info->snd_portid,
 989                                                        info->snd_seq, 0,
 990                                                        OVS_FLOW_CMD_NEW,
 991                                                        ufid_flags);
 992                         BUG_ON(error < 0);
 993                 }
 994                 ovs_unlock();
 995         } else {
 996                 struct sw_flow_actions *old_acts;
 997
 998                 /* Bail out if we're not allowed to modify an existing flow.
 999                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1000                  * because Generic Netlink treats the latter as a dump
1001                  * request.  We also accept NLM_F_EXCL in case that bug ever
1002                  * gets fixed.
1003                  */
1004                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1005                                                          | NLM_F_EXCL))) {
1006                         error = -EEXIST;
1007                         goto err_unlock_ovs;
1008                 }
1009                 /* The flow identifier has to be the same for flow updates.
1010                  * Look for any overlapping flow.
1011                  */
1012                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1013                         if (ovs_identifier_is_key(&flow->id))
1014                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1015                                                                  &match);
1016                         else /* UFID matches but key is different */
1017                                 flow = NULL;
1018                         if (!flow) {
1019                                 error = -ENOENT;
1020                                 goto err_unlock_ovs;
1021                         }
1022                 }
1023                 /* Update actions. */
1024                 old_acts = ovsl_dereference(flow->sf_acts);
1025                 rcu_assign_pointer(flow->sf_acts, acts);
1026
1027                 if (unlikely(reply)) {
1028                         error = ovs_flow_cmd_fill_info(flow,
1029                                                        ovs_header->dp_ifindex,
1030                                                        reply, info->snd_portid,
1031                                                        info->snd_seq, 0,
1032                                                        OVS_FLOW_CMD_NEW,
1033                                                        ufid_flags);
1034                         BUG_ON(error < 0);
1035                 }
1036                 ovs_unlock();
1037
1038                 ovs_nla_free_flow_actions_rcu(old_acts);
1039                 ovs_flow_free(new_flow, false);
1040         }
1041
1042         if (reply)
1043                 ovs_notify(&dp_flow_genl_family, reply, info);
1044         return 0;
1045
1046 err_unlock_ovs:
1047         ovs_unlock();
1048         kfree_skb(reply);
1049 err_kfree_acts:
1050         ovs_nla_free_flow_actions(acts);
1051 err_kfree_flow:
1052         ovs_flow_free(new_flow, false);
1053 error:
1054         return error;
1055 }
1056
1057 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1058 static struct sw_flow_actions *get_flow_actions(struct net *net,
1059                                                 const struct nlattr *a,
1060                                                 const struct sw_flow_key *key,
1061                                                 const struct sw_flow_mask *mask,
1062                                                 bool log)
1063 {
1064         struct sw_flow_actions *acts;
1065         struct sw_flow_key masked_key;
1066         int error;
1067
1068         ovs_flow_mask_key(&masked_key, key, true, mask);
1069         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1070         if (error) {
1071                 OVS_NLERR(log,
1072                           "Actions may not be safe on all matching packets");
1073                 return ERR_PTR(error);
1074         }
1075
1076         return acts;
1077 }
1078
1079 /* Factor out match-init and action-copy to avoid
1080  * "Wframe-larger-than=1024" warning. Because mask is only
1081  * used to get actions, we new a function to save some
1082  * stack space.
1083  *
1084  * If there are not key and action attrs, we return 0
1085  * directly. In the case, the caller will also not use the
1086  * match as before. If there is action attr, we try to get
1087  * actions and save them to *acts. Before returning from
1088  * the function, we reset the match->mask pointer. Because
1089  * we should not to return match object with dangling reference
1090  * to mask.
1091  * */
1092 static int ovs_nla_init_match_and_action(struct net *net,
1093                                          struct sw_flow_match *match,
1094                                          struct sw_flow_key *key,
1095                                          struct nlattr **a,
1096                                          struct sw_flow_actions **acts,
1097                                          bool log)
1098 {
1099         struct sw_flow_mask mask;
1100         int error = 0;
1101
1102         if (a[OVS_FLOW_ATTR_KEY]) {
1103                 ovs_match_init(match, key, true, &mask);
1104                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1105                                           a[OVS_FLOW_ATTR_MASK], log);
1106                 if (error)
1107                         goto error;
1108         }
1109
1110         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1111                 if (!a[OVS_FLOW_ATTR_KEY]) {
1112                         OVS_NLERR(log,
1113                                   "Flow key attribute not present in set flow.");
1114                         return -EINVAL;
1115                 }
1116
1117                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1118                                          &mask, log);
1119                 if (IS_ERR(*acts)) {
1120                         error = PTR_ERR(*acts);
1121                         goto error;
1122                 }
1123         }
1124
1125         /* On success, error is 0. */
1126 error:
1127         match->mask = NULL;
1128         return error;
1129 }
1130
1131 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1132 {
1133         struct net *net = sock_net(skb->sk);
1134         struct nlattr **a = info->attrs;
1135         struct ovs_header *ovs_header = info->userhdr;
1136         struct sw_flow_key key;
1137         struct sw_flow *flow;
1138         struct sk_buff *reply = NULL;
1139         struct datapath *dp;
1140         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1141         struct sw_flow_match match;
1142         struct sw_flow_id sfid;
1143         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1144         int error = 0;
1145         bool log = !a[OVS_FLOW_ATTR_PROBE];
1146         bool ufid_present;
1147
1148         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1149         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1150                 OVS_NLERR(log,
1151                           "Flow set message rejected, Key attribute missing.");
1152                 return -EINVAL;
1153         }
1154
1155         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1156                                               &acts, log);
1157         if (error)
1158                 goto error;
1159
1160         if (acts) {
1161                 /* Can allocate before locking if have acts. */
1162                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1163                                                 ufid_flags);
1164                 if (IS_ERR(reply)) {
1165                         error = PTR_ERR(reply);
1166                         goto err_kfree_acts;
1167                 }
1168         }
1169
1170         ovs_lock();
1171         dp = get_dp(net, ovs_header->dp_ifindex);
1172         if (unlikely(!dp)) {
1173                 error = -ENODEV;
1174                 goto err_unlock_ovs;
1175         }
1176         /* Check that the flow exists. */
1177         if (ufid_present)
1178                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1179         else
1180                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1181         if (unlikely(!flow)) {
1182                 error = -ENOENT;
1183                 goto err_unlock_ovs;
1184         }
1185
1186         /* Update actions, if present. */
1187         if (likely(acts)) {
1188                 old_acts = ovsl_dereference(flow->sf_acts);
1189                 rcu_assign_pointer(flow->sf_acts, acts);
1190
1191                 if (unlikely(reply)) {
1192                         error = ovs_flow_cmd_fill_info(flow,
1193                                                        ovs_header->dp_ifindex,
1194                                                        reply, info->snd_portid,
1195                                                        info->snd_seq, 0,
1196                                                        OVS_FLOW_CMD_NEW,
1197                                                        ufid_flags);
1198                         BUG_ON(error < 0);
1199                 }
1200         } else {
1201                 /* Could not alloc without acts before locking. */
1202                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1203                                                 info, OVS_FLOW_CMD_NEW, false,
1204                                                 ufid_flags);
1205
1206                 if (IS_ERR(reply)) {
1207                         error = PTR_ERR(reply);
1208                         goto err_unlock_ovs;
1209                 }
1210         }
1211
1212         /* Clear stats. */
1213         if (a[OVS_FLOW_ATTR_CLEAR])
1214                 ovs_flow_stats_clear(flow);
1215         ovs_unlock();
1216
1217         if (reply)
1218                 ovs_notify(&dp_flow_genl_family, reply, info);
1219         if (old_acts)
1220                 ovs_nla_free_flow_actions_rcu(old_acts);
1221
1222         return 0;
1223
1224 err_unlock_ovs:
1225         ovs_unlock();
1226         kfree_skb(reply);
1227 err_kfree_acts:
1228         ovs_nla_free_flow_actions(acts);
1229 error:
1230         return error;
1231 }
1232
1233 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1234 {
1235         struct nlattr **a = info->attrs;
1236         struct ovs_header *ovs_header = info->userhdr;
1237         struct net *net = sock_net(skb->sk);
1238         struct sw_flow_key key;
1239         struct sk_buff *reply;
1240         struct sw_flow *flow;
1241         struct datapath *dp;
1242         struct sw_flow_match match;
1243         struct sw_flow_id ufid;
1244         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1245         int err = 0;
1246         bool log = !a[OVS_FLOW_ATTR_PROBE];
1247         bool ufid_present;
1248
1249         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1250         if (a[OVS_FLOW_ATTR_KEY]) {
1251                 ovs_match_init(&match, &key, true, NULL);
1252                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1253                                         log);
1254         } else if (!ufid_present) {
1255                 OVS_NLERR(log,
1256                           "Flow get message rejected, Key attribute missing.");
1257                 err = -EINVAL;
1258         }
1259         if (err)
1260                 return err;
1261
1262         ovs_lock();
1263         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1264         if (!dp) {
1265                 err = -ENODEV;
1266                 goto unlock;
1267         }
1268
1269         if (ufid_present)
1270                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1271         else
1272                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1273         if (!flow) {
1274                 err = -ENOENT;
1275                 goto unlock;
1276         }
1277
1278         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1279                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1280         if (IS_ERR(reply)) {
1281                 err = PTR_ERR(reply);
1282                 goto unlock;
1283         }
1284
1285         ovs_unlock();
1286         return genlmsg_reply(reply, info);
1287 unlock:
1288         ovs_unlock();
1289         return err;
1290 }
1291
1292 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1293 {
1294         struct nlattr **a = info->attrs;
1295         struct ovs_header *ovs_header = info->userhdr;
1296         struct net *net = sock_net(skb->sk);
1297         struct sw_flow_key key;
1298         struct sk_buff *reply;
1299         struct sw_flow *flow = NULL;
1300         struct datapath *dp;
1301         struct sw_flow_match match;
1302         struct sw_flow_id ufid;
1303         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1304         int err;
1305         bool log = !a[OVS_FLOW_ATTR_PROBE];
1306         bool ufid_present;
1307
1308         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1309         if (a[OVS_FLOW_ATTR_KEY]) {
1310                 ovs_match_init(&match, &key, true, NULL);
1311                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1312                                         NULL, log);
1313                 if (unlikely(err))
1314                         return err;
1315         }
1316
1317         ovs_lock();
1318         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1319         if (unlikely(!dp)) {
1320                 err = -ENODEV;
1321                 goto unlock;
1322         }
1323
1324         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1325                 err = ovs_flow_tbl_flush(&dp->table);
1326                 goto unlock;
1327         }
1328
1329         if (ufid_present)
1330                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1331         else
1332                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1333         if (unlikely(!flow)) {
1334                 err = -ENOENT;
1335                 goto unlock;
1336         }
1337
1338         ovs_flow_tbl_remove(&dp->table, flow);
1339         ovs_unlock();
1340
1341         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1342                                         &flow->id, info, false, ufid_flags);
1343         if (likely(reply)) {
1344                 if (likely(!IS_ERR(reply))) {
1345                         rcu_read_lock();        /*To keep RCU checker happy. */
1346                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1347                                                      reply, info->snd_portid,
1348                                                      info->snd_seq, 0,
1349                                                      OVS_FLOW_CMD_DEL,
1350                                                      ufid_flags);
1351                         rcu_read_unlock();
1352                         BUG_ON(err < 0);
1353
1354                         ovs_notify(&dp_flow_genl_family, reply, info);
1355                 } else {
1356                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1357                 }
1358         }
1359
1360         ovs_flow_free(flow, true);
1361         return 0;
1362 unlock:
1363         ovs_unlock();
1364         return err;
1365 }
1366
1367 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1368 {
1369         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1370         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1371         struct table_instance *ti;
1372         struct datapath *dp;
1373         u32 ufid_flags;
1374         int err;
1375
1376         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1377                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1378         if (err)
1379                 return err;
1380         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1381
1382         rcu_read_lock();
1383         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1384         if (!dp) {
1385                 rcu_read_unlock();
1386                 return -ENODEV;
1387         }
1388
1389         ti = rcu_dereference(dp->table.ti);
1390         for (;;) {
1391                 struct sw_flow *flow;
1392                 u32 bucket, obj;
1393
1394                 bucket = cb->args[0];
1395                 obj = cb->args[1];
1396                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1397                 if (!flow)
1398                         break;
1399
1400                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1401                                            NETLINK_CB(cb->skb).portid,
1402                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1403                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1404                         break;
1405
1406                 cb->args[0] = bucket;
1407                 cb->args[1] = obj;
1408         }
1409         rcu_read_unlock();
1410         return skb->len;
1411 }
1412
1413 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1414         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1415         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1416         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1417         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1418         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1419         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1420         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1421 };
1422
1423 static const struct genl_ops dp_flow_genl_ops[] = {
1424         { .cmd = OVS_FLOW_CMD_NEW,
1425           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1426           .policy = flow_policy,
1427           .doit = ovs_flow_cmd_new
1428         },
1429         { .cmd = OVS_FLOW_CMD_DEL,
1430           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431           .policy = flow_policy,
1432           .doit = ovs_flow_cmd_del
1433         },
1434         { .cmd = OVS_FLOW_CMD_GET,
1435           .flags = 0,               /* OK for unprivileged users. */
1436           .policy = flow_policy,
1437           .doit = ovs_flow_cmd_get,
1438           .dumpit = ovs_flow_cmd_dump
1439         },
1440         { .cmd = OVS_FLOW_CMD_SET,
1441           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442           .policy = flow_policy,
1443           .doit = ovs_flow_cmd_set,
1444         },
1445 };
1446
1447 static struct genl_family dp_flow_genl_family __ro_after_init = {
1448         .hdrsize = sizeof(struct ovs_header),
1449         .name = OVS_FLOW_FAMILY,
1450         .version = OVS_FLOW_VERSION,
1451         .maxattr = OVS_FLOW_ATTR_MAX,
1452         .netnsok = true,
1453         .parallel_ops = true,
1454         .ops = dp_flow_genl_ops,
1455         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1456         .mcgrps = &ovs_dp_flow_multicast_group,
1457         .n_mcgrps = 1,
1458         .module = THIS_MODULE,
1459 };
1460
1461 static size_t ovs_dp_cmd_msg_size(void)
1462 {
1463         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1464
1465         msgsize += nla_total_size(IFNAMSIZ);
1466         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1467         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1468         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1469
1470         return msgsize;
1471 }
1472
1473 /* Called with ovs_mutex. */
1474 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1475                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1476 {
1477         struct ovs_header *ovs_header;
1478         struct ovs_dp_stats dp_stats;
1479         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1480         int err;
1481
1482         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1483                                    flags, cmd);
1484         if (!ovs_header)
1485                 goto error;
1486
1487         ovs_header->dp_ifindex = get_dpifindex(dp);
1488
1489         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1490         if (err)
1491                 goto nla_put_failure;
1492
1493         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1494         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1495                           &dp_stats, OVS_DP_ATTR_PAD))
1496                 goto nla_put_failure;
1497
1498         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1499                           sizeof(struct ovs_dp_megaflow_stats),
1500                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1501                 goto nla_put_failure;
1502
1503         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1504                 goto nla_put_failure;
1505
1506         genlmsg_end(skb, ovs_header);
1507         return 0;
1508
1509 nla_put_failure:
1510         genlmsg_cancel(skb, ovs_header);
1511 error:
1512         return -EMSGSIZE;
1513 }
1514
1515 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1516 {
1517         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1518 }
1519
1520 /* Called with rcu_read_lock or ovs_mutex. */
1521 static struct datapath *lookup_datapath(struct net *net,
1522                                         const struct ovs_header *ovs_header,
1523                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1524 {
1525         struct datapath *dp;
1526
1527         if (!a[OVS_DP_ATTR_NAME])
1528                 dp = get_dp(net, ovs_header->dp_ifindex);
1529         else {
1530                 struct vport *vport;
1531
1532                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1533                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1534         }
1535         return dp ? dp : ERR_PTR(-ENODEV);
1536 }
1537
1538 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1539 {
1540         struct datapath *dp;
1541
1542         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1543         if (IS_ERR(dp))
1544                 return;
1545
1546         WARN(dp->user_features, "Dropping previously announced user features\n");
1547         dp->user_features = 0;
1548 }
1549
1550 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1551 {
1552         if (a[OVS_DP_ATTR_USER_FEATURES])
1553                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1554 }
1555
1556 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1557 {
1558         struct nlattr **a = info->attrs;
1559         struct vport_parms parms;
1560         struct sk_buff *reply;
1561         struct datapath *dp;
1562         struct vport *vport;
1563         struct ovs_net *ovs_net;
1564         int err, i;
1565
1566         err = -EINVAL;
1567         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1568                 goto err;
1569
1570         reply = ovs_dp_cmd_alloc_info();
1571         if (!reply)
1572                 return -ENOMEM;
1573
1574         err = -ENOMEM;
1575         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1576         if (dp == NULL)
1577                 goto err_free_reply;
1578
1579         ovs_dp_set_net(dp, sock_net(skb->sk));
1580
1581         /* Allocate table. */
1582         err = ovs_flow_tbl_init(&dp->table);
1583         if (err)
1584                 goto err_free_dp;
1585
1586         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1587         if (!dp->stats_percpu) {
1588                 err = -ENOMEM;
1589                 goto err_destroy_table;
1590         }
1591
1592         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1593                             GFP_KERNEL);
1594         if (!dp->ports) {
1595                 err = -ENOMEM;
1596                 goto err_destroy_percpu;
1597         }
1598
1599         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1600                 INIT_HLIST_HEAD(&dp->ports[i]);
1601
1602         /* Set up our datapath device. */
1603         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1604         parms.type = OVS_VPORT_TYPE_INTERNAL;
1605         parms.options = NULL;
1606         parms.dp = dp;
1607         parms.port_no = OVSP_LOCAL;
1608         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1609
1610         ovs_dp_change(dp, a);
1611
1612         /* So far only local changes have been made, now need the lock. */
1613         ovs_lock();
1614
1615         vport = new_vport(&parms);
1616         if (IS_ERR(vport)) {
1617                 err = PTR_ERR(vport);
1618                 if (err == -EBUSY)
1619                         err = -EEXIST;
1620
1621                 if (err == -EEXIST) {
1622                         /* An outdated user space instance that does not understand
1623                          * the concept of user_features has attempted to create a new
1624                          * datapath and is likely to reuse it. Drop all user features.
1625                          */
1626                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1627                                 ovs_dp_reset_user_features(skb, info);
1628                 }
1629
1630                 goto err_destroy_ports_array;
1631         }
1632
1633         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1634                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1635         BUG_ON(err < 0);
1636
1637         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1638         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1639
1640         ovs_unlock();
1641
1642         ovs_notify(&dp_datapath_genl_family, reply, info);
1643         return 0;
1644
1645 err_destroy_ports_array:
1646         ovs_unlock();
1647         kfree(dp->ports);
1648 err_destroy_percpu:
1649         free_percpu(dp->stats_percpu);
1650 err_destroy_table:
1651         ovs_flow_tbl_destroy(&dp->table);
1652 err_free_dp:
1653         kfree(dp);
1654 err_free_reply:
1655         kfree_skb(reply);
1656 err:
1657         return err;
1658 }
1659
1660 /* Called with ovs_mutex. */
1661 static void __dp_destroy(struct datapath *dp)
1662 {
1663         int i;
1664
1665         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1666                 struct vport *vport;
1667                 struct hlist_node *n;
1668
1669                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1670                         if (vport->port_no != OVSP_LOCAL)
1671                                 ovs_dp_detach_port(vport);
1672         }
1673
1674         list_del_rcu(&dp->list_node);
1675
1676         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1677          * all ports in datapath are destroyed first before freeing datapath.
1678          */
1679         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1680
1681         /* RCU destroy the flow table */
1682         call_rcu(&dp->rcu, destroy_dp_rcu);
1683 }
1684
1685 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1686 {
1687         struct sk_buff *reply;
1688         struct datapath *dp;
1689         int err;
1690
1691         reply = ovs_dp_cmd_alloc_info();
1692         if (!reply)
1693                 return -ENOMEM;
1694
1695         ovs_lock();
1696         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1697         err = PTR_ERR(dp);
1698         if (IS_ERR(dp))
1699                 goto err_unlock_free;
1700
1701         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1702                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1703         BUG_ON(err < 0);
1704
1705         __dp_destroy(dp);
1706         ovs_unlock();
1707
1708         ovs_notify(&dp_datapath_genl_family, reply, info);
1709
1710         return 0;
1711
1712 err_unlock_free:
1713         ovs_unlock();
1714         kfree_skb(reply);
1715         return err;
1716 }
1717
1718 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1719 {
1720         struct sk_buff *reply;
1721         struct datapath *dp;
1722         int err;
1723
1724         reply = ovs_dp_cmd_alloc_info();
1725         if (!reply)
1726                 return -ENOMEM;
1727
1728         ovs_lock();
1729         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1730         err = PTR_ERR(dp);
1731         if (IS_ERR(dp))
1732                 goto err_unlock_free;
1733
1734         ovs_dp_change(dp, info->attrs);
1735
1736         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1737                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1738         BUG_ON(err < 0);
1739
1740         ovs_unlock();
1741         ovs_notify(&dp_datapath_genl_family, reply, info);
1742
1743         return 0;
1744
1745 err_unlock_free:
1746         ovs_unlock();
1747         kfree_skb(reply);
1748         return err;
1749 }
1750
1751 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1752 {
1753         struct sk_buff *reply;
1754         struct datapath *dp;
1755         int err;
1756
1757         reply = ovs_dp_cmd_alloc_info();
1758         if (!reply)
1759                 return -ENOMEM;
1760
1761         ovs_lock();
1762         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1763         if (IS_ERR(dp)) {
1764                 err = PTR_ERR(dp);
1765                 goto err_unlock_free;
1766         }
1767         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1768                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1769         BUG_ON(err < 0);
1770         ovs_unlock();
1771
1772         return genlmsg_reply(reply, info);
1773
1774 err_unlock_free:
1775         ovs_unlock();
1776         kfree_skb(reply);
1777         return err;
1778 }
1779
1780 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1781 {
1782         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1783         struct datapath *dp;
1784         int skip = cb->args[0];
1785         int i = 0;
1786
1787         ovs_lock();
1788         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1789                 if (i >= skip &&
1790                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1791                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1792                                          OVS_DP_CMD_NEW) < 0)
1793                         break;
1794                 i++;
1795         }
1796         ovs_unlock();
1797
1798         cb->args[0] = i;
1799
1800         return skb->len;
1801 }
1802
1803 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1804         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1805         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1806         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1807 };
1808
1809 static const struct genl_ops dp_datapath_genl_ops[] = {
1810         { .cmd = OVS_DP_CMD_NEW,
1811           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1812           .policy = datapath_policy,
1813           .doit = ovs_dp_cmd_new
1814         },
1815         { .cmd = OVS_DP_CMD_DEL,
1816           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1817           .policy = datapath_policy,
1818           .doit = ovs_dp_cmd_del
1819         },
1820         { .cmd = OVS_DP_CMD_GET,
1821           .flags = 0,               /* OK for unprivileged users. */
1822           .policy = datapath_policy,
1823           .doit = ovs_dp_cmd_get,
1824           .dumpit = ovs_dp_cmd_dump
1825         },
1826         { .cmd = OVS_DP_CMD_SET,
1827           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1828           .policy = datapath_policy,
1829           .doit = ovs_dp_cmd_set,
1830         },
1831 };
1832
1833 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1834         .hdrsize = sizeof(struct ovs_header),
1835         .name = OVS_DATAPATH_FAMILY,
1836         .version = OVS_DATAPATH_VERSION,
1837         .maxattr = OVS_DP_ATTR_MAX,
1838         .netnsok = true,
1839         .parallel_ops = true,
1840         .ops = dp_datapath_genl_ops,
1841         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1842         .mcgrps = &ovs_dp_datapath_multicast_group,
1843         .n_mcgrps = 1,
1844         .module = THIS_MODULE,
1845 };
1846
1847 /* Called with ovs_mutex or RCU read lock. */
1848 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1849                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1850 {
1851         struct ovs_header *ovs_header;
1852         struct ovs_vport_stats vport_stats;
1853         int err;
1854
1855         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1856                                  flags, cmd);
1857         if (!ovs_header)
1858                 return -EMSGSIZE;
1859
1860         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1861
1862         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1863             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1864             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1865                            ovs_vport_name(vport)))
1866                 goto nla_put_failure;
1867
1868         ovs_vport_get_stats(vport, &vport_stats);
1869         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1870                           sizeof(struct ovs_vport_stats), &vport_stats,
1871                           OVS_VPORT_ATTR_PAD))
1872                 goto nla_put_failure;
1873
1874         if (ovs_vport_get_upcall_portids(vport, skb))
1875                 goto nla_put_failure;
1876
1877         err = ovs_vport_get_options(vport, skb);
1878         if (err == -EMSGSIZE)
1879                 goto error;
1880
1881         genlmsg_end(skb, ovs_header);
1882         return 0;
1883
1884 nla_put_failure:
1885         err = -EMSGSIZE;
1886 error:
1887         genlmsg_cancel(skb, ovs_header);
1888         return err;
1889 }
1890
1891 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1892 {
1893         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1894 }
1895
1896 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1897 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1898                                          u32 seq, u8 cmd)
1899 {
1900         struct sk_buff *skb;
1901         int retval;
1902
1903         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1904         if (!skb)
1905                 return ERR_PTR(-ENOMEM);
1906
1907         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1908         BUG_ON(retval < 0);
1909
1910         return skb;
1911 }
1912
1913 /* Called with ovs_mutex or RCU read lock. */
1914 static struct vport *lookup_vport(struct net *net,
1915                                   const struct ovs_header *ovs_header,
1916                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1917 {
1918         struct datapath *dp;
1919         struct vport *vport;
1920
1921         if (a[OVS_VPORT_ATTR_NAME]) {
1922                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1923                 if (!vport)
1924                         return ERR_PTR(-ENODEV);
1925                 if (ovs_header->dp_ifindex &&
1926                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1927                         return ERR_PTR(-ENODEV);
1928                 return vport;
1929         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1930                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1931
1932                 if (port_no >= DP_MAX_PORTS)
1933                         return ERR_PTR(-EFBIG);
1934
1935                 dp = get_dp(net, ovs_header->dp_ifindex);
1936                 if (!dp)
1937                         return ERR_PTR(-ENODEV);
1938
1939                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1940                 if (!vport)
1941                         return ERR_PTR(-ENODEV);
1942                 return vport;
1943         } else
1944                 return ERR_PTR(-EINVAL);
1945 }
1946
1947 /* Called with ovs_mutex */
1948 static void update_headroom(struct datapath *dp)
1949 {
1950         unsigned dev_headroom, max_headroom = 0;
1951         struct net_device *dev;
1952         struct vport *vport;
1953         int i;
1954
1955         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1956                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1957                         dev = vport->dev;
1958                         dev_headroom = netdev_get_fwd_headroom(dev);
1959                         if (dev_headroom > max_headroom)
1960                                 max_headroom = dev_headroom;
1961                 }
1962         }
1963
1964         dp->max_headroom = max_headroom;
1965         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1966                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1967                         netdev_set_rx_headroom(vport->dev, max_headroom);
1968 }
1969
1970 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1971 {
1972         struct nlattr **a = info->attrs;
1973         struct ovs_header *ovs_header = info->userhdr;
1974         struct vport_parms parms;
1975         struct sk_buff *reply;
1976         struct vport *vport;
1977         struct datapath *dp;
1978         u32 port_no;
1979         int err;
1980
1981         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1982             !a[OVS_VPORT_ATTR_UPCALL_PID])
1983                 return -EINVAL;
1984
1985         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1986                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1987         if (port_no >= DP_MAX_PORTS)
1988                 return -EFBIG;
1989
1990         reply = ovs_vport_cmd_alloc_info();
1991         if (!reply)
1992                 return -ENOMEM;
1993
1994         ovs_lock();
1995 restart:
1996         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1997         err = -ENODEV;
1998         if (!dp)
1999                 goto exit_unlock_free;
2000
2001         if (port_no) {
2002                 vport = ovs_vport_ovsl(dp, port_no);
2003                 err = -EBUSY;
2004                 if (vport)
2005                         goto exit_unlock_free;
2006         } else {
2007                 for (port_no = 1; ; port_no++) {
2008                         if (port_no >= DP_MAX_PORTS) {
2009                                 err = -EFBIG;
2010                                 goto exit_unlock_free;
2011                         }
2012                         vport = ovs_vport_ovsl(dp, port_no);
2013                         if (!vport)
2014                                 break;
2015                 }
2016         }
2017
2018         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2019         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2020         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2021         parms.dp = dp;
2022         parms.port_no = port_no;
2023         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2024
2025         vport = new_vport(&parms);
2026         err = PTR_ERR(vport);
2027         if (IS_ERR(vport)) {
2028                 if (err == -EAGAIN)
2029                         goto restart;
2030                 goto exit_unlock_free;
2031         }
2032
2033         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2034                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2035
2036         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2037                 update_headroom(dp);
2038         else
2039                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2040
2041         BUG_ON(err < 0);
2042         ovs_unlock();
2043
2044         ovs_notify(&dp_vport_genl_family, reply, info);
2045         return 0;
2046
2047 exit_unlock_free:
2048         ovs_unlock();
2049         kfree_skb(reply);
2050         return err;
2051 }
2052
2053 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2054 {
2055         struct nlattr **a = info->attrs;
2056         struct sk_buff *reply;
2057         struct vport *vport;
2058         int err;
2059
2060         reply = ovs_vport_cmd_alloc_info();
2061         if (!reply)
2062                 return -ENOMEM;
2063
2064         ovs_lock();
2065         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2066         err = PTR_ERR(vport);
2067         if (IS_ERR(vport))
2068                 goto exit_unlock_free;
2069
2070         if (a[OVS_VPORT_ATTR_TYPE] &&
2071             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2072                 err = -EINVAL;
2073                 goto exit_unlock_free;
2074         }
2075
2076         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2077                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2078                 if (err)
2079                         goto exit_unlock_free;
2080         }
2081
2082
2083         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2084                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2085
2086                 err = ovs_vport_set_upcall_portids(vport, ids);
2087                 if (err)
2088                         goto exit_unlock_free;
2089         }
2090
2091         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2092                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2093         BUG_ON(err < 0);
2094
2095         ovs_unlock();
2096         ovs_notify(&dp_vport_genl_family, reply, info);
2097         return 0;
2098
2099 exit_unlock_free:
2100         ovs_unlock();
2101         kfree_skb(reply);
2102         return err;
2103 }
2104
2105 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2106 {
2107         bool must_update_headroom = false;
2108         struct nlattr **a = info->attrs;
2109         struct sk_buff *reply;
2110         struct datapath *dp;
2111         struct vport *vport;
2112         int err;
2113
2114         reply = ovs_vport_cmd_alloc_info();
2115         if (!reply)
2116                 return -ENOMEM;
2117
2118         ovs_lock();
2119         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2120         err = PTR_ERR(vport);
2121         if (IS_ERR(vport))
2122                 goto exit_unlock_free;
2123
2124         if (vport->port_no == OVSP_LOCAL) {
2125                 err = -EINVAL;
2126                 goto exit_unlock_free;
2127         }
2128
2129         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2130                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2131         BUG_ON(err < 0);
2132
2133         /* the vport deletion may trigger dp headroom update */
2134         dp = vport->dp;
2135         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2136                 must_update_headroom = true;
2137         netdev_reset_rx_headroom(vport->dev);
2138         ovs_dp_detach_port(vport);
2139
2140         if (must_update_headroom)
2141                 update_headroom(dp);
2142         ovs_unlock();
2143
2144         ovs_notify(&dp_vport_genl_family, reply, info);
2145         return 0;
2146
2147 exit_unlock_free:
2148         ovs_unlock();
2149         kfree_skb(reply);
2150         return err;
2151 }
2152
2153 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2154 {
2155         struct nlattr **a = info->attrs;
2156         struct ovs_header *ovs_header = info->userhdr;
2157         struct sk_buff *reply;
2158         struct vport *vport;
2159         int err;
2160
2161         reply = ovs_vport_cmd_alloc_info();
2162         if (!reply)
2163                 return -ENOMEM;
2164
2165         rcu_read_lock();
2166         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2167         err = PTR_ERR(vport);
2168         if (IS_ERR(vport))
2169                 goto exit_unlock_free;
2170         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2171                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2172         BUG_ON(err < 0);
2173         rcu_read_unlock();
2174
2175         return genlmsg_reply(reply, info);
2176
2177 exit_unlock_free:
2178         rcu_read_unlock();
2179         kfree_skb(reply);
2180         return err;
2181 }
2182
2183 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2184 {
2185         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2186         struct datapath *dp;
2187         int bucket = cb->args[0], skip = cb->args[1];
2188         int i, j = 0;
2189
2190         rcu_read_lock();
2191         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2192         if (!dp) {
2193                 rcu_read_unlock();
2194                 return -ENODEV;
2195         }
2196         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2197                 struct vport *vport;
2198
2199                 j = 0;
2200                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2201                         if (j >= skip &&
2202                             ovs_vport_cmd_fill_info(vport, skb,
2203                                                     NETLINK_CB(cb->skb).portid,
2204                                                     cb->nlh->nlmsg_seq,
2205                                                     NLM_F_MULTI,
2206                                                     OVS_VPORT_CMD_NEW) < 0)
2207                                 goto out;
2208
2209                         j++;
2210                 }
2211                 skip = 0;
2212         }
2213 out:
2214         rcu_read_unlock();
2215
2216         cb->args[0] = i;
2217         cb->args[1] = j;
2218
2219         return skb->len;
2220 }
2221
2222 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2223         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2224         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2225         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2226         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2227         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2228         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2229 };
2230
2231 static const struct genl_ops dp_vport_genl_ops[] = {
2232         { .cmd = OVS_VPORT_CMD_NEW,
2233           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2234           .policy = vport_policy,
2235           .doit = ovs_vport_cmd_new
2236         },
2237         { .cmd = OVS_VPORT_CMD_DEL,
2238           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2239           .policy = vport_policy,
2240           .doit = ovs_vport_cmd_del
2241         },
2242         { .cmd = OVS_VPORT_CMD_GET,
2243           .flags = 0,               /* OK for unprivileged users. */
2244           .policy = vport_policy,
2245           .doit = ovs_vport_cmd_get,
2246           .dumpit = ovs_vport_cmd_dump
2247         },
2248         { .cmd = OVS_VPORT_CMD_SET,
2249           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2250           .policy = vport_policy,
2251           .doit = ovs_vport_cmd_set,
2252         },
2253 };
2254
2255 struct genl_family dp_vport_genl_family __ro_after_init = {
2256         .hdrsize = sizeof(struct ovs_header),
2257         .name = OVS_VPORT_FAMILY,
2258         .version = OVS_VPORT_VERSION,
2259         .maxattr = OVS_VPORT_ATTR_MAX,
2260         .netnsok = true,
2261         .parallel_ops = true,
2262         .ops = dp_vport_genl_ops,
2263         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2264         .mcgrps = &ovs_dp_vport_multicast_group,
2265         .n_mcgrps = 1,
2266         .module = THIS_MODULE,
2267 };
2268
2269 static struct genl_family * const dp_genl_families[] = {
2270         &dp_datapath_genl_family,
2271         &dp_vport_genl_family,
2272         &dp_flow_genl_family,
2273         &dp_packet_genl_family,
2274 };
2275
2276 static void dp_unregister_genl(int n_families)
2277 {
2278         int i;
2279
2280         for (i = 0; i < n_families; i++)
2281                 genl_unregister_family(dp_genl_families[i]);
2282 }
2283
2284 static int __init dp_register_genl(void)
2285 {
2286         int err;
2287         int i;
2288
2289         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2290
2291                 err = genl_register_family(dp_genl_families[i]);
2292                 if (err)
2293                         goto error;
2294         }
2295
2296         return 0;
2297
2298 error:
2299         dp_unregister_genl(i);
2300         return err;
2301 }
2302
2303 static int __net_init ovs_init_net(struct net *net)
2304 {
2305         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2306
2307         INIT_LIST_HEAD(&ovs_net->dps);
2308         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2309         ovs_ct_init(net);
2310         return 0;
2311 }
2312
2313 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2314                                             struct list_head *head)
2315 {
2316         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2317         struct datapath *dp;
2318
2319         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2320                 int i;
2321
2322                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2323                         struct vport *vport;
2324
2325                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2326                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2327                                         continue;
2328
2329                                 if (dev_net(vport->dev) == dnet)
2330                                         list_add(&vport->detach_list, head);
2331                         }
2332                 }
2333         }
2334 }
2335
2336 static void __net_exit ovs_exit_net(struct net *dnet)
2337 {
2338         struct datapath *dp, *dp_next;
2339         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2340         struct vport *vport, *vport_next;
2341         struct net *net;
2342         LIST_HEAD(head);
2343
2344         ovs_ct_exit(dnet);
2345         ovs_lock();
2346         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2347                 __dp_destroy(dp);
2348
2349         rtnl_lock();
2350         for_each_net(net)
2351                 list_vports_from_net(net, dnet, &head);
2352         rtnl_unlock();
2353
2354         /* Detach all vports from given namespace. */
2355         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2356                 list_del(&vport->detach_list);
2357                 ovs_dp_detach_port(vport);
2358         }
2359
2360         ovs_unlock();
2361
2362         cancel_work_sync(&ovs_net->dp_notify_work);
2363 }
2364
2365 static struct pernet_operations ovs_net_ops = {
2366         .init = ovs_init_net,
2367         .exit = ovs_exit_net,
2368         .id   = &ovs_net_id,
2369         .size = sizeof(struct ovs_net),
2370 };
2371
2372 static int __init dp_init(void)
2373 {
2374         int err;
2375
2376         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2377
2378         pr_info("Open vSwitch switching datapath\n");
2379
2380         err = action_fifos_init();
2381         if (err)
2382                 goto error;
2383
2384         err = ovs_internal_dev_rtnl_link_register();
2385         if (err)
2386                 goto error_action_fifos_exit;
2387
2388         err = ovs_flow_init();
2389         if (err)
2390                 goto error_unreg_rtnl_link;
2391
2392         err = ovs_vport_init();
2393         if (err)
2394                 goto error_flow_exit;
2395
2396         err = register_pernet_device(&ovs_net_ops);
2397         if (err)
2398                 goto error_vport_exit;
2399
2400         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2401         if (err)
2402                 goto error_netns_exit;
2403
2404         err = ovs_netdev_init();
2405         if (err)
2406                 goto error_unreg_notifier;
2407
2408         err = dp_register_genl();
2409         if (err < 0)
2410                 goto error_unreg_netdev;
2411
2412         return 0;
2413
2414 error_unreg_netdev:
2415         ovs_netdev_exit();
2416 error_unreg_notifier:
2417         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2418 error_netns_exit:
2419         unregister_pernet_device(&ovs_net_ops);
2420 error_vport_exit:
2421         ovs_vport_exit();
2422 error_flow_exit:
2423         ovs_flow_exit();
2424 error_unreg_rtnl_link:
2425         ovs_internal_dev_rtnl_link_unregister();
2426 error_action_fifos_exit:
2427         action_fifos_exit();
2428 error:
2429         return err;
2430 }
2431
2432 static void dp_cleanup(void)
2433 {
2434         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2435         ovs_netdev_exit();
2436         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2437         unregister_pernet_device(&ovs_net_ops);
2438         rcu_barrier();
2439         ovs_vport_exit();
2440         ovs_flow_exit();
2441         ovs_internal_dev_rtnl_link_unregister();
2442         action_fifos_exit();
2443 }
2444
2445 module_init(dp_init);
2446 module_exit(dp_cleanup);
2447
2448 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2449 MODULE_LICENSE("GPL");
2450 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2451 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2452 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2453 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);